RAVEN: Radar Adaptive Vision Encoders for Efficient Chirp-wise Object Detection and Segmentation
This addresses the problem of high computational cost and latency in radar perception for autonomous vehicles, representing an incremental improvement over existing radar pipelines.
The paper tackles efficient object detection and segmentation from FMCW radar data by proposing RAVEN, a deep learning architecture that processes raw ADC data chirp-wise and uses an early-exit mechanism. The result is strong performance on automotive radar benchmarks with substantial reductions in computation and latency compared to conventional frame-based methods.
This paper presents RAVEN, a computationally efficient deep learning architecture for FMCW radar perception. The method processes raw ADC data in a chirp-wise streaming manner, preserves MIMO structure through independent receiver state-space encoders, and uses a learnable cross-antenna mixing module to recover compact virtual-array features. It also introduces an early-exit mechanism so the model can make decisions using only a subset of chirps when the latent state has stabilized. Across automotive radar benchmarks, the approach reports strong object detection and BEV free-space segmentation performance while substantially reducing computation and end-to-end latency compared with conventional frame-based radar pipelines.